BibTeX Library Subset Builder

How to use BibTeX Library Subset Builder

Share the relevant part of an owned bibliography without retyping its TeX fields. Enter exact citation keys and create a smaller .bib file from the original record slices. A selected entry’s plain crossref parent is retained as well; all string and preamble directives keep their source order. The evidence export explains which records were kept and where they came from.

1. Choose your local UTF-8 .bib file or paste its source. Enter 1–100 distinct exact citation keys on separate lines; use source case.
2. Run the extraction. Review added crossref parents, omitted entry counts and the warning that every string and preamble directive remains.
3. Download the .bib source or its JSON evidence. Review retained metadata and TeX before sharing, then test the file with your own bibliography application and style.

Good to know

• This selected source profile accepts the fourteen classic entry types and @string/@preamble directives, with braces or parentheses around records. A trailing comma may follow an entry field list, but cannot follow a string or preamble value. It refuses @comment, unknown/extended entry types and duplicate fields. Free material outside records is omitted. Percent is ordinary content inside a quoted/braced value; it does not start a field comment.
• Citation keys use 1–128 ASCII characters: first letter, digit or underscore, followed by letters, digits, underscore, period, colon, plus, slash or hyphen. Exact duplicate or case-only colliding library/requested keys refuse. Keys are not inferred from manuscripts, .aux files, URLs or titles. Missing requested keys refuse the entire export.
• Values consist of balanced braced/quoted text, digit strings or ASCII named macros, optionally joined by #. At most 32 tokens per value and 32 inner brace levels. Backslash does not escape a BibTeX quote or brace for this scanner: TeX accent quotes belong inside balanced braces. Raw TeX, Unicode, number spelling and inner whitespace remain unchanged; macros and names are not expanded or interpreted.
• Only selected one-level crossref relationships are assessed. Each selected crossref must be one braced/quoted plain key; its parent must exist later in the original source and cannot itself crossref. Selected xdata/xref/related/entryset/ids fields refuse. Other omitted-entry relationships and TeX cite commands inside fields are not followed. Parent source retention does not instruct BibTeX to cite the parent.
• All @string and @preamble records are retained at their original relative positions, without pruning, expansion or validity evaluation. They may contain unrelated private metadata or TeX code. This is not anonymization, sanitization, full BibTeX/BibLaTeX validation, citation formatting, publication verification or a TeX execution environment. Never compile untrusted TeX without an appropriate separate environment.
• Use at most 500 entries, 200 string directives, 10 preambles and 10,000 entry fields. Source is at most 200,000 UTF-16 units and 600,000 UTF-8 bytes; a real file must be nonempty strict UTF-8. At most one initial BOM is accepted and omitted from output. Malformed Unicode, unsupported controls and further BOMs refuse.
• Each retained record is copied exactly. Separators become two LF characters and the result ends with LF, so outside comments/spacing and BOM bytes are not preserved. Source output is at most 210,000 units and 700,000 bytes; JSON evidence at most 1,500,000 bytes. Offsets are zero-based UTF-16 positions in the original decoded source, inclusive start and exclusive end, including an initial BOM if present.
• Processing stays local, without uploads, network requests, storage or user code execution. A downloaded source file is not proof of compatibility with every reference manager, .bst style or older non-UTF-8 BibTeX installation. Actual browser download, reset and CSP checks remain separate release gates.